Replacing tokens when computing completion results is uncertain

[fm-117]

When computing completion proposals, more precisely in CodeElementMatcher.MatchCompletionCodeElement, the server may replace the closest token to cursor with a fake UserDefinedWord. This is prone to error as the text from this made-up token is later used to filter proposals and if it was non-alphanumeric, the completion won't find anything.

As far as our unit tests can tell, this behavior is used to prevent completion from returning anything when invoked after a single : or a .: see CompletionAfterBetweenColons and CompletionAfterDotFromTypedDataDef LSR tests.

Removing the token replacing code has no effect except for these 2 tests so we should look for a way to replicate current behavior without having to create those fake tokens.

TypeCobol/TypeCobol.LanguageServer/Completion Factory/CodeElementMatcher.cs

Lines 93 to 109 in 930ed19

	//The closestToken to cursor as to be added to this codeElement as a UserDefinedWord
	if (closestTokenToCursor.TokenType != TokenType.UserDefinedWord)
	{
	var codeElement = codeElements.LastOrDefault();
	if (codeElement != null)
	{
	//As we are altering our input, keep track of changes for later restore
	codeElementsArrangedTokensToRestore.Add(new Tuple<CodeElementWrapper, List<Token>>(codeElement, codeElement.ArrangedConsumedTokens));
	codeElement.ArrangedConsumedTokens = codeElement.ArrangedConsumedTokens.ConvertAll(ReplaceClosestTokenToCursor);
	}
	Token ReplaceClosestTokenToCursor(Token originalToken)
	{
	return closestTokenToCursor.Equals(originalToken)
	? new Token(TokenType.UserDefinedWord, closestTokenToCursor.StartIndex, closestTokenToCursor.StopIndex, closestTokenToCursor.TokensLine)
	: originalToken;
	}